基于K-means聚类的多智能体跟随多领导者算法

为防止多智能体集群跟随多个领导者时编队混乱，提出了3种K-means聚类算法，将集群分成与领导者数量相同的社区，社区内的智能体将跟随同一领导者。所提出的3种算法中，算法1适用于智能体分布空间广的场景，系统达到一致性所需时间最短；算法2则适用于智能体分布稀疏的场景，可有效避免智能体碰撞等危险；算法3则大大降低多智能体集群的控制成本，但将会牺牲系统的收敛速度。相较于传统预先对智能体编号，领导-跟随关系固定的方法，本文提出的分簇方法使系统收敛时间更短，且有效应对中途任务变更的情况，可快速分配给智能体新的合适任务。     

无人机集群牵制控制系统特征值分析

本文研究施加牵制控制对无人机(unmanned aerial vehicle,UAV)集群蜂拥控制一致性的影响.首先基于复杂网络和图论建立了UAV集群的二阶动力学模型;其次设计了分布式控制协议并构建了集群系统的增广Laplacian矩阵,其最小特征值反映了集群的一致性收敛速度,以及连通性与鲁棒性;最后分析了增加牵制点对...     

基于均匀设计的多智能体遗传算法研究

将均匀设计方法、多智能体系统与遗传算法相结合,提出了一种多智能体遗传算法.设计了基于均匀设计表的均匀种群初始化方法和均匀交叉算子,并将其引入算法中.算法通过智能体与智能体环境间的相互作用,以及智能体自学习提高自身能量,实现了复杂问题的全局优化目标.实验表明,该算法对于高维函数优化问题,尤其是高维多峰函数,具有很好的全局搜索能力和较快的收敛速度,能够避免陷入局部最优.     

基于MAS的关系网络交易成本分析仿真模型

针对关系网络交易成本研究方法在随机性、自主性和整体涌现性表现的不足,分析了关系网络中企业交易过程及交易成本产生的原因,提出了基于多智能体系统的关系网络交易成本分析仿真模型。该模型利用多智能体系统解决了自主性和整体涌现性问题,利用仿真技术解决了随机性问题。然后采用microsoft.net平台开发了本仿真模型的软件系统,应用于算例,得出了几个关系网络中交易成本的重要结论。可扩展本模型进行关系网络的其他方面的研究。     

时滞多智能体系统关于参考状态的信息一致性

研究了具有参考状态的有向通信时延多智能体系统的信息一致性问题.提出了适用于可连续通信并有通信时延的有向多智能体系统关于参考状态的一致性算法,严格推导了多智能体系统在该算法作用下达到信息一致性的充分性判据,最后以水下无人航行器协同作战为背景,仿真验证了所提出的一致性算法和判据的有效性.     

基于CAS理论的多智能体战斗模型仿真研究

为了研究基于复杂适应系统理论的多智能体仿真模型的演化、涌现规律,在Swarm平台上建立了一个多智能体的战斗模拟仿真系统,并对战斗过程双方Agent数量和实力的演化进行了模拟仿真,得到了和传统数学模型不一致的结果.基于复杂适应系统理论的多主体仿真模型可以模拟复杂系统的微观行为,进而能够显现出系统宏观涌现现象,非常适合模拟复杂的战争系统.     

复杂任务下的多智能体协同进化算法

针对多智能体系统在处理复杂任务时存在的低效率、高冗积、多智能体系统内协同模型算法存在交互冲突、资源损耗过高等问题,提出一种基于复杂任务的多智能体系统优化算法.在差分进化算法与局部优化算法的基础上对二者进行优化,结合强化学习的训练框架,构建训练网络,通过对学习步长进行修订,改变种群迭代优化准则,使得种群在计算力充足的情况...     

单边Lipschitz非线性多智能体系统一致性追踪控制

针对单边Lipschitz非线性多智能体系统,提出了一种分布式一致性控制方法.首先,构建了领导-跟随者动力学结构,用于实现单边Lipschitz多智能体系统的追踪控制.然后,设计了单边Lipschitz非线性多智能体系统的一致性控制协议,可根据智能体之间局部交互信息构建分布式反馈控制,并将系统的一致性追踪问题转化为系统...     

基于改进势场的无人机编队恢复与一致性仿真

针对多无人机编队防撞、编队恢复及位置和速度收敛一致性问题,提出以改进势场原理与一致性理论为基础的分布式协同编队控制算法。建立静态障碍物模型、无人机质点模型与二阶系统动态模型;定义含协调因子和通信权重的机间协调势场函数,实现防撞和队形恢复的控制目标;在基本一致性协议中引入编队中心参考向量、期望速度和速度镇定项,实现位置和速度的收敛一致性;结合改进的一致性协议与改进势场作用下的合加速度设计协同编队控制算法,并由Hamilton函数证明该算法的稳定收敛性。仿真结果表明：该方法可以实现防撞、队形恢复及位置与速度收敛一致的控制目标。     

拓扑可调无标度网络上的雪堆博弈研究

复杂网络上博弈个体间合作行为的涌现具有重要的现实意义.为了研究无标度网络拓扑结构对合作涌现的影响,结合雪堆博弈模型和比例模仿策略,研究了拓扑结构可调的无标度网络上的博弈动力学演化,具体研究内容为网络的幂律指数、平均度和平均聚类系数与合作密度的关系.仿真结果表明,无标度网络的合作密度与网络度分布的均匀程度正相关,高聚类结构可以有效增强网络上的合作密度,即幂律指数越高,平均聚类系数越小时,合作水平越低;另外,平均度与合作密度呈现非单调关系,存在一个最佳平均度.     

多智能体系统的有限时间跟踪控制

基于领导-跟随者模型研究了多智能体系统的有限时间跟踪控制问题。针对领导者状态是时变的情况,提出了一类非线性有限时间跟踪控制算法,在固定网络拓扑结构下,利用Lyapunov有限时间稳定性理论和矩阵理论分析,得到了该算法使得系统中跟随者状态在有限时间内与领导者状态达到一致的充分条件;在切换网络拓扑结构下,提出了一类有限时间跟踪控制算法,在领导者的状态是时不变的情况下,给出了该算法使得系统实现有限时间跟踪控制的充分条件。仿真实例验证了所提出两类算法的有效性。     

OPTIMAL TARGET TRAJECTORY ESTIMATION AND FILTERING USING NETWORKED SENSORS

Target tracking using distributed sensor network is in general a challenging problem because it always needs to deal with real-time processing of noisy information. In this paper the problem of using nonlinear sensors such as distance and direction sensors for estimating a moving target is studied. The problem is formulated as a prudent design of nonlinear filters for a linear system subject to noisy nonlinear measurements and partially unknown input, which is generated by an exogenous system. In the worst case where the input is completely unknown, the exogenous dynamics is reduced to the random walk model. It can be shown that the nonlinear filter will have optimal convergence if the number of the sensors are large enough and the convergence rate will be highly improved if the sensors are deployed appropriately. This actually raises an interesting issue on active sensing： how to optimally move the sensors if they are considered as mobile multi-agent systems？ Finally, a simulation example is given to illustrate and validate the construction of our filter.     

A quantum multi-agent based neural network model for failure prediction

An effective prognostic program is crucial to the predictive maintenance of complex equipment since it can improve productivity, prolong equipment life, and enhance system safety. This paper proposes a novel technique for accurate failure prognosis based on back propagation neural network and quantum multi-agent algorithm. Inspired by the extensive research of quantum computing theory and multi-agent systems, the technique employs a quantum multi-agent strategy, with the main characteristics of quantum agent representation and several operations including fitness evaluation, cooperation, crossover and mutation, for parameters optimization of neural network to avoid the deficiencies such as slow convergence and liability of getting stuck to local minima. To validate the feasibility of the proposed approach, several numerical approximation experiments were firstly designed, after which real vibrational data of bearings from the Laboratory of Cincinnati University were analyzed and used to assess the health condition for a given future point. The results were rather encouraging and indicated that the presented forecasting method has the potential to be utilized as an estimation tool for failure prediction in industrial machinery.     

基于多Agent系统网管模型的形式化建模

运用多Agent系统实现分布、自治的网络管理是构造下一代网络管理系统的一条全新思路。然而,大多数基于多Agent系统的网管模型均采用非形式化的方法进行描述,容易造成人们对模型理解的歧义性和在系统开发过程中的随意性。一种利用有色Petri网对分布式网络管理模型进行形式化建模的方法被提出。利用该方法所建立的形式化模型,能够对分布式网管模型中多Agent系统的组成结构和运行机制进行全面刻画,能够描述模型中用到的Agent的各种特性。     

Consensus for Heterogeneous Multi-Agent Systems with Directed Network Topologies

In this paper, the consensus problem for heterogeneous multi-agent systems composed of first-order and second-order agents is investigated with directed network topologies. Based on a system transformation method, this consensus problem is turned into a consensus problem for homogeneous multi-agent systems. With certain assumption on the control parameters, firstly, necessary and sufficient condition for consensus is proposed with fixed topology. Secondly, sufficient condition is proposed for heterogeneous multi-agent systems to achieve consensus with switching topologies. Finally, simulation examples are presented to verify the effectiveness of the theoretical results.     

DIDSS环境下信息agent系统任务规范的分解

提出一个基于GDF(扩展决策函数)的多agent任务规范分解判据，提出多agent的任务规范分解性质；将多agent任务规范分解问题等价为建立一个与信息agent任务规范属性相关的MSBN(多模块贝叶斯网)问题，并证明了MSBN的d-cutset(d-割集)和分解判据之间的关系，提出了与多agent任务规范对应的MSBN的分解方法．最后通过示例验证了所提出的方法的可行性。     

Sliding Mode Formation Control of Nonlinear Multi-agent Systems with Local Lipschitz Continuous Dynamics

In this paper, the multi-agent formation problem of networked nonlinear multi-agent systems with local Lipschitz continuous dynamics under directed interaction topology, is investigated.Based on the nonlinear dynamics satisfying locally Lipschitz continuous conditions, three kinds of sliding mode controllers are proposed to solve the problem of multi-agent formation control. Using integral sliding mode controller in first-order system, formation shape is achieved within finite time.For second-order system, on the one hand, non-singular terminal sliding mode function is adopted to accomplish the system asymptotic convergence. Furthermore, super-twisting algorithm is proposed to make multi-agent achieve the desired formation within finite time. Lyapunov functions are applied in the whole paper to ensure the system stability. Numerical simulation examples are provided to demonstrate the effectiveness of the proposed sliding mode control methods.     

具有不同输入时延的二阶多智能体系统一致性

研究了具有不同时变输入时延的二阶连续多智能体系统的一致性问题。首先,通过变量转换,将系统的收敛性问题转化为误差系统的稳定问题;然后,通过对系统进行变换,将二阶系统稳定性问题转换为等价系统的稳定性问题。通过构造李雅普诺夫函数,基于线性矩阵不等式(LMI)的方法,给出在无向固定拓扑条件下,系统达到一致的充分条件。最后,仿真实例证明了结果的有效性。     

Bipartite Consensus of Linear Multi-Agent Systems by Distributed Event-Triggered Control

For multi-agent systems with competitive and collaborative relationships, signed graph can more intuitively express the characteristics of their interactive networks. In this paper, the bipartite consensus is investigated for multi-agent systems with structurally balanced signed graph. In order to reduce actuation burden in dynamical network environment, the event-triggering strategy is applied to bipartite consensus protocol for the multi-agent systems. The triggered condition for each agent is designed by using its own information and transmitted information of its neighbors at sampling instant and make the number of triggers of the whole systems be reduced. Based on the distributed eventtriggered control, some sufficient conditions are derived to guarantee the leaderless and leader-following bipartite consensus. Finally, some numerical examples are shown to demonstrate the effectiveness of the theoretical results.     

基于博弈论的多传感器交叉提示算法

针对传感器交叉提示技术在目标探测中的应用问题展开研究,首先介绍多传感器交叉提示技术基本理论,建立了具有社区性的传感器网络模型和基于多Agent技术的分布式传感器管理结构,分析了传感器交叉提示发生的条件和类型,其次提出了传感器网络效能评价函数,最后设计了基于博弈论的多传感器交叉提示算法。仿真结果表明,提出的多传感器交叉提示模型符合实际作战场景,能够生成传感器联盟方案对目标检测和跟踪,并进行目标交接,与拍卖算法和人工蜂群算法相比,博弈论算法求解质量较高,收敛速度明显提高,更适用于作战态势较紧急的情形。